The U.S. NIC Operational MIZ product classifies sea ice into two categories; between 10% and 80% sea ice concentration and greater than 80% sea ice concentration. This data is initially saved in shapefile format, then converted to ASCII, KMZ and png formats. The U.S. NIC daily graphics page generally describes the process used to generate this product;
The daily ice edge is analyzed by sea ice experts using multiple sources of near real time satellite data, derived satellite products, buoy data, weather, and analyst interpretation of current sea ice conditions.
Historical png images can be found here and all other data types can be found here. Data from this product spans from 2010-current.

The NSIDC offers a climate data record of sea ice concentration derived from passive microwave data. This dataset is generated from two commonly-used passive microwave algorithms then merged together based on the algorithm with the higher value. This data set spans 1987-current, however only the 2010-current data that overlaps with the NIC product was used in this comparison. The CDR archive can be found here and additional information on this product can be found with the user guide here.

The NIC product user guide draft makes an important note that differentiates this product from the NSIDC dataset;
Mariners would not expect to encounter sea ice in significant concentrations when sailing outside of the line. Therefore, while it can be thought of as a 10% contour, on a practical level it functions as a contour between open ocean and ice at any concentration.
Therefore, the outer edge of the MIZ as shown by the US NIC product will likely be farther out to sea than the NSIDC product, since the lowest sea ice concentration measured by the latter is 15%.
Per the NSIDC user guide on the CDR product;
...the concentration value given by the NASA Team algorithm and that given by the Bootstrap algorithm are compared; whichever value is greater is selected as the CDR value. This is done because both algorithms tend to underestimate ice concentration, however the source of this bias differs between algorithms (Meier et al. 2014).
If the passive microwave algorithms do typically underestimate the sea ice concentration values, then the same sea ice concentration thresholds for the inner and outer boundary of the MIZ would result in an MIZ that is further inland when compared to the NIC product.
The following workflow will help compare these two products to get a better sense of which product has a greater extent and, if the extents differ, when do they differ the most? To start with, the workflow will download all raw data from NSIDC servers between the provided start and end date. It will then convert this data into comparable numpy arrays and save them to disk so this workflow can be re-run easily with different sea ice concentration thresholds. For the CDR data, this workflow will simply extract the sea ice concentration variable and write this out to disk. For the NIC shapefiles, the shapes are rasterized to a grid with the same extent as the CDR data. The rasterization process marks any pixel whose center falls within the shape the value of the shape. Using an all-touched method, where any pixels that are touched by a shape are categorized as within the shape, resulted in a significant overestimate of the sea ice extent.
When running the workflow over a large period of time (in the images populated below, the entire 10-year history was run), it takes a very long time to complete the download and conversion process (it typically takes about 6 hours to download and convert all the data). Once the data is downloaded and optimized, the workflow should run in around 5 minutes.
This workflow is set up to run the same comparison for both the southern hemisphere and the northern hemisphere. The northern hemisphere takes roughly twice as long to optimize since there are more shapes to be rasterized in the NIC data.
After downloading and converting the data, this workflow runs through the following analysis steps;
These start and end times are configurable and referenced throughout this notebook - this is the time range you would like to analyze. Note that longer time periods take a much longer time to analyze (specifically in the downloading and rasterization steps). Discussion and interpretation in this notebook is based on the variables as set below.
The following calls download and convert data. The execution of this cell will take a long time the first time around and should be faster thereafter.
Let's take an initial look at the NIC data. A few things to note about this data;
Let's take an initial look at the CDR data. A few things to note about this data;
Plotted below are the monthly median sea ice extent for the 80% threshold for each month in the last few years. Note the sea ice extends significantly farther out to sea in the NIC interpretation. This is especially evident in the springtime months when melting accelerates and the ice edge recedes. Passive Microwave models typically underestimate sea ice concentration, so this type of bias is somewhat expected.
Also, the edge of this extent is significantly smoother than the edge drawn in the NIC dataset. This may be due to spatial smoothing in the CDR algorithm that is a result of how this data is measured. Ice measurement values in each pixel in the CDR product can contaminate neighboring pixels, resulting in this type of spatial smoothing.
This will likely be the longest-running cell of the notebook - it creates a dataframe that stores total area of sea ice within varying sea ice concentrations - from 5% to 95% - for both products.
Now, let's take a closer look at the actual ice areas within the outer and inner marginal ice zone boundaries for the two products. The plot below shows the MIZ area over the last several years for both the CDR product and the NIC product. Note you can see the outer and inner edge variation is present spatially (eg we can see a more jagged edge in the diagram above) as well as temporally (eg we see a jagged line in the magenta time series but a smoother line in the cyan time series below).
The plot below also shows the seasonal relationship between these two products - the difference betwen the two areas is greatest when ice is melting. Additionally, the right axis represents the total area of the marginal ice zone. It's interesting to note that, while the CDR product generally represents the marginal ice zone as further receeded towards the south, the total size of the marginal ice zone is typically larger than that of the NIC product.
Finally, I was interested in seeing what the longer-term relationship between these two sets of data looked like. After overlaying annual data (February 15th-February 15th) for the entire history of the NIC product, the NIC perimeters were consistently farther north than the CDR perimeters.
Now that we've confirmed the CDR product does typically underestimate the sea ice concentration when compared to the NIC product, I'm hoping to quantify exactly what this overestimate looks like. In the plot below, I've taken the mean differences between the 80% NIC threshold and varying CDR thresholds over the entire history of data. I've found that CDR thresholds of 30% sea ice concentration typically line up with NIC thresholds of 80%.
Additionally, I've plotted out these different CDR thresholds over the recent past. The 30% CDR threshold does visually line up here as well.
I've selected the 30% CDR threshold and reproduced the graph above. The lower boundaries of these areas do roughly line up with each other. Note the upper boundaries, which represent the outer threshold of the marginal ice zone, represent different sea ice concentrations and have not been changed from their original values.
This further visually confirms the 30% threshold for the CDR data generally agrees with the historical 80% NIC sea ice concentration threshold as well.
Another method of comparing these two datasets is by comparing how closely they agree with each other based on the footprints of the MIZ. One method to do this, which is outlined fairly clearly by Liu et al, is by generating a contingency table. This contingency table quantifies how often;
From this information, you can determine a contingency table for each daily grid (see below for the letters that line up with the code variables A/B/C/D).
| CDR MIZ | CDR no MIZ | |
|---|---|---|
| NIC MIZ | "A" | "C" |
| NIC no MIZ | "B" | "D" |
From this contingency table, you can then calculate a Hanssen-Kuiper Skill Score (KSS), reviewed in Liu et al as;
$Hit = \frac{A}{A + C}$
$False Alarm Rate = \frac{B}{B + D}$
$KSS = Hit - False Alarm Rate$
Below, I have calculated and plotted the KSS over time. Unfortunately, time limitations for this project will push further analysis of this into next steps.
I've updated the median sea ice extent plots to use the 30% sea ice concentration threshold for the CDR dataset and found that the two datasets do line up well when viewing the perimeters on a monthly basis as well.
Meier, W. N. 2005. Comparison of Passive Microwave Ice Concentration Algorithm Retrievals with AVHRR Imagery in Arctic Peripheral Seas. IEEE Trans. Geosci. Remote Sens., 43(6): 1324-1337.
Liu, Y., Helfrich S., Meier W.N., Dworak R. 2020. Assessment of AMSR2 Ice Extent and Ice Edge in the Arctic Using IMS. Remote Sensing.